Configuring cabal to tell HPC to ignore certain functions

[ivanperez-keera]

I want to configure a cabal package (specifically, copilot-core) so that, when HPC is executed during testing with coverage enabled, it ignores certain definitions, for example:

• Anything that is auto-generated (e.g., Show).
• Data constructors and field accessors, for which GHC also defines functions automatically, are all assumed to work.
• Proxy doesn't need to be WHNFed.

I think there's a way to extract the tix file and modify it, but I'm trying to make this automated, and configure the cabal package so that the right options are passed by cabal to HPC, and that way even hackage will pick it up when it reports the coverage of our library.

Any pointers?

Also posted at: https://www.reddit.com/r/haskell/comments/1g8mm82/configuring_cabal_to_tell_hpc_to_ignore_certain/

[ivanperez-keera]

Follow-up: I managed to get over 90%, but I'm still finding it hard to hit some expressions. My current branch is: https://github.com/ivanperez-keera/copilot/tree/develop-copilot-core-coverage.

This is what I'm seeing:

Some cases are due to laziness:

In other cases, I don't understand why I can't hit the expression. For example:

I thought this would be explored fully during comparison of values generated using typeOf using equality, but it suggests that it never looks at the typeOf inside an array.

Help to improve the tests to increase coverage would be appreciated.

[RyanGlScott]

I think the reason that the typeOf argument to Array is never evaluated is because Array stores types in two ways:

data Type :: * -> * where
  Array :: forall n t. (KnownNat n, Typed t) => Type t -> Type (Array n t)

Namely, it stores both an implicit Typed t constraint as well as an explicit Type t field. It appears that the surrounding code is only making use of the Typed t constraint without making use of the Type t field.

One proposal: what if we got rid of either the Typed t constraint or the Type t field? I claim that having both is redundant, as you can always convert from a Typed t constraint to a Type t value (and vice versa). My belief is that the coverage issue related to the Typed (Array n t) instance would go away if we did this.

[RyanGlScott]

Actually, there might be a simpler solution. Are you certain that the copilot-core test suite actually makes meaningful use of the Typed (Array n t) instance? I suspect that the call sites to typeOf are only evaluating the Array data constructor, but not the underlying field. For instance, if I make the following quick-and-dirty change to the copilot-core test suite:

diff --git a/copilot-core/tests/Test/Copilot/Core/Type.hs b/copilot-core/tests/Test/Copilot/Core/Type.hs
index 9f9cb6d5..ccce150d 100644
--- a/copilot-core/tests/Test/Copilot/Core/Type.hs
+++ b/copilot-core/tests/Test/Copilot/Core/Type.hs
@@ -253,7 +253,7 @@ utypes =
     ]
   where
     a1 :: Type (Array 3 Int8)
-    a1 = Array Int8
+    a1 = typeOf
 
     a2 :: Type (Array 4 Int8)
     a2 = Array Int8

Then I do see the typeOf in typeOf = Array typeOf become evaluated in the coverage report.

[ivanperez-keera]

There is now an issue in HPC to track this: https://gitlab.haskell.org/hpc/hpc-bin/-/issues/21